**Fetal Health Classification**

GOAL:

Create a model for**Fetal Health Classification,** Classify the health of a fetus as Normal, Suspect or Pathological using CTG data

1. Obtain the dataset with 2126 records of features extracted from Cardiotocogram (CTG) exams.

#import python packages
import pandas as pd
import numpy as np
import random
from sklearn.preprocessing import LabelEncoder

from sklearn.model_selection import train_test_split

#Load dataset
fetal_health=pd.read_csv('fetal_health.csv')

fetal_health.head(3)

	baseline value	accelerations	fetal_movement	uterine_contractions	light_decelerations	severe_decelerations	prolongued_decelerations	abnormal_short_term_variability	mean_value_of_short_term_variability	percentage_of_time_with_abnormal_long_term_variability	...	histogram_min	histogram_max	histogram_number_of_peaks	histogram_number_of_zeroes	histogram_mode	histogram_mean	histogram_median	histogram_variance	histogram_tendency	fetal_health
0	120.0	0.000	0.0	0.000	0.000	0.0	0.0	73.0	0.5	43.0	...	62.0	126.0	2.0	0.0	120.0	137.0	121.0	73.0	1.0	2.0
1	132.0	0.006	0.0	0.006	0.003	0.0	0.0	17.0	2.1	0.0	...	68.0	198.0	6.0	1.0	141.0	136.0	140.0	12.0	0.0	1.0
2	133.0	0.003	0.0	0.008	0.003	0.0	0.0	16.0	2.1	0.0	...	68.0	198.0	5.0	1.0	141.0	135.0	138.0	13.0	0.0	1.0

3 rows × 22 columns

#Check dtype of data
fetal_health.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 2126 entries, 0 to 2125
Data columns (total 22 columns):
 #   Column                                                  Non-Null Count  Dtype  
---  ------                                                  --------------  -----  
 0   baseline value                                          2126 non-null   float64
 1   accelerations                                           2126 non-null   float64
 2   fetal_movement                                          2126 non-null   float64
 3   uterine_contractions                                    2126 non-null   float64
 4   light_decelerations                                     2126 non-null   float64
 5   severe_decelerations                                    2126 non-null   float64
 6   prolongued_decelerations                                2126 non-null   float64
 7   abnormal_short_term_variability                         2126 non-null   float64
 8   mean_value_of_short_term_variability                    2126 non-null   float64
 9   percentage_of_time_with_abnormal_long_term_variability  2126 non-null   float64
 10  mean_value_of_long_term_variability                     2126 non-null   float64
 11  histogram_width                                         2126 non-null   float64
 12  histogram_min                                           2126 non-null   float64
 13  histogram_max                                           2126 non-null   float64
 14  histogram_number_of_peaks                               2126 non-null   float64
 15  histogram_number_of_zeroes                              2126 non-null   float64
 16  histogram_mode                                          2126 non-null   float64
 17  histogram_mean                                          2126 non-null   float64
 18  histogram_median                                        2126 non-null   float64
 19  histogram_variance                                      2126 non-null   float64
 20  histogram_tendency                                      2126 non-null   float64
 21  fetal_health                                            2126 non-null   float64
dtypes: float64(22)
memory usage: 365.5 KB

fetal_health.shape

(2126, 22)

fetal_health.dtypes

baseline value                                            float64
accelerations                                             float64
fetal_movement                                            float64
uterine_contractions                                      float64
light_decelerations                                       float64
severe_decelerations                                      float64
prolongued_decelerations                                  float64
abnormal_short_term_variability                           float64
mean_value_of_short_term_variability                      float64
percentage_of_time_with_abnormal_long_term_variability    float64
mean_value_of_long_term_variability                       float64
histogram_width                                           float64
histogram_min                                             float64
histogram_max                                             float64
histogram_number_of_peaks                                 float64
histogram_number_of_zeroes                                float64
histogram_mode                                            float64
histogram_mean                                            float64
histogram_median                                          float64
histogram_variance                                        float64
histogram_tendency                                        float64
fetal_health                                              float64
dtype: object

#Check null value for each column
fetal_health.isna().any()

baseline value                                            False
accelerations                                             False
fetal_movement                                            False
uterine_contractions                                      False
light_decelerations                                       False
severe_decelerations                                      False
prolongued_decelerations                                  False
abnormal_short_term_variability                           False
mean_value_of_short_term_variability                      False
percentage_of_time_with_abnormal_long_term_variability    False
mean_value_of_long_term_variability                       False
histogram_width                                           False
histogram_min                                             False
histogram_max                                             False
histogram_number_of_peaks                                 False
histogram_number_of_zeroes                                False
histogram_mode                                            False
histogram_mean                                            False
histogram_median                                          False
histogram_variance                                        False
histogram_tendency                                        False
fetal_health                                              False
dtype: bool

fetal_health.describe()

	baseline value	accelerations	fetal_movement	uterine_contractions	light_decelerations	severe_decelerations	prolongued_decelerations	abnormal_short_term_variability	mean_value_of_short_term_variability	percentage_of_time_with_abnormal_long_term_variability	...	histogram_min	histogram_max	histogram_number_of_peaks	histogram_number_of_zeroes	histogram_mode	histogram_mean	histogram_median	histogram_variance	histogram_tendency	fetal_health
count	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.00000	...	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000	2126.000000
mean	133.303857	0.003178	0.009481	0.004366	0.001889	0.000003	0.000159	46.990122	1.332785	9.84666	...	93.579492	164.025400	4.068203	0.323612	137.452023	134.610536	138.090310	18.808090	0.320320	1.304327
std	9.840844	0.003866	0.046666	0.002946	0.002960	0.000057	0.000590	17.192814	0.883241	18.39688	...	29.560212	17.944183	2.949386	0.706059	16.381289	15.593596	14.466589	28.977636	0.610829	0.614377
min	106.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	12.000000	0.200000	0.00000	...	50.000000	122.000000	0.000000	0.000000	60.000000	73.000000	77.000000	0.000000	-1.000000	1.000000
25%	126.000000	0.000000	0.000000	0.002000	0.000000	0.000000	0.000000	32.000000	0.700000	0.00000	...	67.000000	152.000000	2.000000	0.000000	129.000000	125.000000	129.000000	2.000000	0.000000	1.000000
50%	133.000000	0.002000	0.000000	0.004000	0.000000	0.000000	0.000000	49.000000	1.200000	0.00000	...	93.000000	162.000000	3.000000	0.000000	139.000000	136.000000	139.000000	7.000000	0.000000	1.000000
75%	140.000000	0.006000	0.003000	0.007000	0.003000	0.000000	0.000000	61.000000	1.700000	11.00000	...	120.000000	174.000000	6.000000	0.000000	148.000000	145.000000	148.000000	24.000000	1.000000	1.000000
max	160.000000	0.019000	0.481000	0.015000	0.015000	0.001000	0.005000	87.000000	7.000000	91.00000	...	159.000000	238.000000	18.000000	10.000000	187.000000	182.000000	186.000000	269.000000	1.000000	3.000000

8 rows × 22 columns

Upon examining the dataset, it appears to be exceptionally clean. There are no null values present in any of the columns, and all the data is in a numerical format.

2. EDA

# Load python packages
import matplotlib.pyplot as plt
import seaborn as sns

# Distribution of target variable (fetal health)
plt.figure(figsize=(8,6))
sns.countplot(x='fetal_health', data=fetal_health)
plt.title('Fetal Health Distribution')
plt.xlabel('Fetal Health')
plt.show()

# Count the number of instances in each class
class_counts = fetal_health['fetal_health'].value_counts()

# Print the class distribution
print('Class distribution:')
print(class_counts)

Class distribution:
fetal_health
1.0    1655
2.0     295
3.0     176
Name: count, dtype: int64

#Correlation Heatmap
plt.figure(figsize=(12,12))
sns.heatmap(fetal_health.corr(),annot=True,cmap='coolwarm',vmin=-1,vmax=1)
plt.title('Correlation Heatmap')

Text(0.5, 1.0, 'Correlation Heatmap')

# Pairplot for visualizing relationships and distribution
sns.pairplot(fetal_health,hue='fetal_health', diag_kind='hist', palette='Set2')
plt.show()

# Box plots for each feature grouped by fetal health
Fetal_list=fetal_health.columns.tolist()
Fetal_list.remove('fetal_health')
Fetal_list

['baseline value',
 'accelerations',
 'fetal_movement',
 'uterine_contractions',
 'light_decelerations',
 'severe_decelerations',
 'prolongued_decelerations',
 'abnormal_short_term_variability',
 'mean_value_of_short_term_variability',
 'percentage_of_time_with_abnormal_long_term_variability',
 'mean_value_of_long_term_variability',
 'histogram_width',
 'histogram_min',
 'histogram_max',
 'histogram_number_of_peaks',
 'histogram_number_of_zeroes',
 'histogram_mode',
 'histogram_mean',
 'histogram_median',
 'histogram_variance',
 'histogram_tendency']

for feature in Fetal_list:
    plt.figure(figsize=(4,3))
    sns.boxenplot(x='fetal_health',y=feature,data=fetal_health,palette='Set2')
    plt.title(f'{feature} vs Fetal Health')
    plt.xlabel('Fetal Health')
    plt.ylabel(feature)
    plt.show()

I want to try a new package called autoviz.classify_method

from autoviz.classify_method import data_cleaning_suggestions ,data_suggestions

Imported v0.1.601. After importing, execute '%matplotlib inline' to display charts in Jupyter.
    AV = AutoViz_Class()
    dfte = AV.AutoViz(filename, sep=',', depVar='', dfte=None, header=0, verbose=1, lowess=False,
               chart_format='svg',max_rows_analyzed=150000,max_cols_analyzed=30, save_plot_dir=None)
Update: verbose=0 displays charts in your local Jupyter notebook.
        verbose=1 additionally provides EDA data cleaning suggestions. It also displays charts.
        verbose=2 does not display charts but saves them in AutoViz_Plots folder in local machine.
        chart_format='bokeh' displays charts in your local Jupyter notebook.
        chart_format='server' displays charts in your browser: one tab for each chart type
        chart_format='html' silently saves interactive HTML files in your local machine

#data cleaning suggestion
data_cleaning_suggestions(fetal_health)

Data cleaning improvement suggestions. Complete them before proceeding to ML modeling.

	Nullpercent	NuniquePercent	dtype	Nuniques	Nulls	Least num. of categories	Data cleaning improvement suggestions
baseline value	0.000000	2.257761	float64	48	0	0
accelerations	0.000000	0.940734	float64	20	0	0	right skewed distribution: cap or drop outliers
histogram_tendency	0.000000	0.141110	float64	3	0	0
histogram_variance	0.000000	6.255880	float64	133	0	0	right skewed distribution: cap or drop outliers
histogram_median	0.000000	4.468485	float64	95	0	0
histogram_mean	0.000000	4.844779	float64	103	0	0
histogram_mode	0.000000	4.139229	float64	88	0	0	left skewed distribution: cap or drop outliers
histogram_number_of_zeroes	0.000000	0.423330	float64	9	0	0	right skewed distribution: cap or drop outliers
histogram_number_of_peaks	0.000000	0.846660	float64	18	0	0
histogram_max	0.000000	4.045155	float64	86	0	0
histogram_min	0.000000	5.126999	float64	109	0	0
histogram_width	0.000000	7.243650	float64	154	0	0
mean_value_of_long_term_variability	0.000000	11.712135	float64	249	0	0	right skewed distribution: cap or drop outliers
percentage_of_time_with_abnormal_long_term_variability	0.000000	4.092192	float64	87	0	0	right skewed distribution: cap or drop outliers
mean_value_of_short_term_variability	0.000000	2.681091	float64	57	0	0	right skewed distribution: cap or drop outliers
abnormal_short_term_variability	0.000000	3.527752	float64	75	0	0
prolongued_decelerations	0.000000	0.282220	float64	6	0	0	right skewed distribution: cap or drop outliers
severe_decelerations	0.000000	0.094073	float64	2	0	0	highly right skewed distribution: drop outliers or do box-cox transform
light_decelerations	0.000000	0.752587	float64	16	0	0	right skewed distribution: cap or drop outliers
uterine_contractions	0.000000	0.752587	float64	16	0	0
fetal_movement	0.000000	4.797742	float64	102	0	0	highly right skewed distribution: drop outliers or do box-cox transform
fetal_health	0.000000	0.141110	float64	3	0	0	right skewed distribution: cap or drop outliers

# I will use 'SMOTE' to figure out this situation. the SMOTE module provides the Synthetic Minority Over-sampling Technique (SMOTE) algorithm for generating synthetic samples of the minority class in an imbalanced dataset.
from imblearn.over_sampling import SMOTE

X=fetal_health[Fetal_list]
y=fetal_health['fetal_health']

X_train,X_val,y_train,y_val=train_test_split(X,y,test_size=0.2,random_state=42)

# use 'SMOTE' modify x and y
# Instantiate SMOTE
sm = SMOTE(random_state=42)

# Fit SMOTE to training data
X_train_res, y_train_res = sm.fit_resample(X_train, y_train)

# Print class distribution of original and resampled data
print('Class distribution before resampling:')
print(y_train.value_counts())
print('Class distribution after resampling:')
print(y_train_res.value_counts())

Class distribution before resampling:
fetal_health
1.0    1322
2.0     231
3.0     147
Name: count, dtype: int64
Class distribution after resampling:
fetal_health
1.0    1322
2.0    1322
3.0    1322
Name: count, dtype: int64

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, confusion_matrix, accuracy_score

# Standardizing the data (scaling)
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train_res)
X_test_scaled = scaler.transform(X_val)

4. Develop a multiclass classification model using relevant algorithms (e.g., logistic regression, decision trees, or deep learning).

#Load Packages
from pycaret  import regression
from sklearn.model_selection import cross_val_score

from sklearn.linear_model import LogisticRegression
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import AdaBoostClassifier
from sklearn.metrics import confusion_matrix
from sklearn.metrics import ConfusionMatrixDisplay

Train_data=pd.concat([X_train_res, y_train_res], axis=1)
#Train_data.head(2)

from pycaret.classification import *
s = setup(data=Train_data, target='fetal_health', session_id=123, normalize=True)

	Description	Value
0	Session id	123
1	Target	fetal_health
2	Target type	Multiclass
3	Target mapping	1.0: 0, 2.0: 1, 3.0: 2
4	Original data shape	(3966, 22)
5	Transformed data shape	(3966, 22)
6	Transformed train set shape	(2776, 22)
7	Transformed test set shape	(1190, 22)
8	Numeric features	21
9	Preprocess	True
10	Imputation type	simple
11	Numeric imputation	mean
12	Categorical imputation	mode
13	Normalize	True
14	Normalize method	zscore
15	Fold Generator	StratifiedKFold
16	Fold Number	10
17	CPU Jobs	-1
18	Use GPU	False
19	Log Experiment	False
20	Experiment Name	clf-default-name
21	USI	cba7

compare_models()

Initiated	. . . . . . . . . . . . . . . . . .	20:38:18
Status	. . . . . . . . . . . . . . . . . .	Loading Dependencies
Estimator	. . . . . . . . . . . . . . . . . .	Compiling Library

LGBMClassifier(boosting_type='gbdt', class_weight=None, colsample_bytree=1.0,
               importance_type='split', learning_rate=0.1, max_depth=-1,
               min_child_samples=20, min_child_weight=0.001, min_split_gain=0.0,
               n_estimators=100, n_jobs=-1, num_leaves=31, objective=None,
               random_state=123, reg_alpha=0.0, reg_lambda=0.0, silent='warn',
               subsample=1.0, subsample_for_bin=200000, subsample_freq=0)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Try LightGBM, RandomForest, KNN

# Compare the performance of several classification models
best_model = compare_models()

# Train the best model on the full dataset
trained_model = create_model(best_model)

# Print the detailed performance metrics for the trained model
print(trained_model)

The best model is :LGBMClassifier(boosting_type='gbdt', class_weight=None, colsample_bytree=1.0, importance_type='split', learning_rate=0.1, max_depth=-1, min_child_samples=20, min_child_weight=0.001, min_split_gain=0.0, n_estimators=100, n_jobs=-1, num_leaves=31, objective=None, random_state=123, reg_alpha=0.0, reg_lambda=0.0, silent='warn', subsample=1.0, subsample_for_bin=200000, subsample_freq=0)

from sklearn.model_selection import cross_val_score
lgbm=trained_model
# Evaluate the ensemble model using cross-validation
scores = cross_val_score(lgbm, X_train_res, y_train_res, cv=20)

preds = predict_model(lgbm)

	Model	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC
0	Light Gradient Boosting Machine	0.9832	0.9985	0.9937	0.9790	0.9863	0.9748	0.9749

#EDA modle
plot_model(lgbm)

plot_model(trained_model, plot = 'error')

plot_model(lgbm, plot = 'learning')

plot_model(lgbm, plot = 'vc')

print("Accuracy of lgbmBoost:",lgbm.score(X_val, y_val))

Accuracy of lgbmBoost: 0.15023474178403756

# Random Forest
import pycaret.classification as clf

rf = create_model('rf')

Initiated	. . . . . . . . . . . . . . . . . .	17:05:46
Status	. . . . . . . . . . . . . . . . . .	Loading Dependencies
Estimator	. . . . . . . . . . . . . . . . . .	Compiling Library

	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC
Fold
0	0.9676	0.9984	0.9676	0.9681	0.9677	0.9514	0.9516
1	0.9604	0.9982	0.9604	0.9617	0.9606	0.9406	0.9412
2	0.9748	0.9978	0.9748	0.9755	0.9749	0.9622	0.9625
3	0.9748	0.9993	0.9748	0.9754	0.9749	0.9622	0.9625
4	0.9892	0.9997	0.9892	0.9894	0.9892	0.9838	0.9839
5	0.9568	0.9956	0.9568	0.9568	0.9566	0.9353	0.9355
6	0.9783	0.9993	0.9783	0.9783	0.9783	0.9675	0.9675
7	0.9567	0.9971	0.9567	0.9579	0.9563	0.9350	0.9360
8	0.9711	0.9987	0.9711	0.9711	0.9711	0.9567	0.9567
9	0.9783	0.9984	0.9783	0.9790	0.9785	0.9675	0.9677
Mean	0.9708	0.9982	0.9708	0.9713	0.9708	0.9562	0.9565
Std	0.0100	0.0011	0.0100	0.0098	0.0101	0.0150	0.0148

# Evaluate the ensemble model using cross-validation
rf_scores = cross_val_score(rf, X_train_res, y_train_res, cv=20)

preds_rf = predict_model(rf)

	Model	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC
0	Random Forest Classifier	0.9824	0.9989	0.9912	0.9802	0.9856	0.9735	0.9736

#EDA modle
plot_model(rf)

plot_model(rf, plot = 'error')

plot_model(rf, plot = 'learning')

plot_model(rf, plot = 'vc')

print("Accuracy of Random Forest:",rf.score(X_val, y_val))

Accuracy of Random Forest: 0.704225352112676

y_pred_rf = rf.predict(X_val)

# Evaluating the model
print("Confusion Matrix:")
print(confusion_matrix(y_val, y_pred_rf))
print("\nClassification Report:")
print(classification_report(y_val, y_pred_rf))
print("Accuracy Score:")
print(accuracy_score(y_val, y_pred_rf))

Confusion Matrix:
[[239  94   0]
 [  3  61   0]
 [ 17  12   0]]

Classification Report:
              precision    recall  f1-score   support

         1.0       0.92      0.72      0.81       333
         2.0       0.37      0.95      0.53        64
         3.0       0.00      0.00      0.00        29

    accuracy                           0.70       426
   macro avg       0.43      0.56      0.45       426
weighted avg       0.78      0.70      0.71       426

Accuracy Score:
0.704225352112676

# KNN

knn  = KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
                     metric_params=None, n_jobs=-1, n_neighbors=5, p=2,
                     weights='uniform')
knn.fit(X_train_res, y_train_res)

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
                     metric_params=None, n_jobs=-1, n_neighbors=5, p=2,
                     weights='uniform')

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

print(knn)

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
                     metric_params=None, n_jobs=-1, n_neighbors=5, p=2,
                     weights='uniform')

# Evaluate the ensemble model using cross-validation
knn_scores = cross_val_score(knn, X_train_res, y_train_res, cv=20)

#preds_knn = predict_model(knn)

#EDA modle
plot_model(knn)

#plot_model(knn, plot = 'error')

plot_model(knn, plot = 'learning')

plot_model(knn, plot = 'vc')

y_pred_knn = knn.predict(X_val)

# Evaluating the model
print("Confusion Matrix:")
print(confusion_matrix(y_val, y_pred_knn))
print("\nClassification Report:")
print(classification_report(y_val, y_pred_knn))
print("Accuracy Score:")
print(accuracy_score(y_val, y_pred_knn))

Confusion Matrix:
[[  0   0   0   0]
 [303  30   0   0]
 [ 26  38   0   0]
 [ 15  12   2   0]]

Classification Report:
              precision    recall  f1-score   support

         0.0       0.00      0.00      0.00         0
         1.0       0.38      0.09      0.15       333
         2.0       0.00      0.00      0.00        64
         3.0       0.00      0.00      0.00        29

    accuracy                           0.07       426
   macro avg       0.09      0.02      0.04       426
weighted avg       0.29      0.07      0.11       426

Accuracy Score:
0.07042253521126761

I have tested three models, namely KNN, Random Forest and LightGBM. We can see that the best model is random forest, which has good effects both in accuracy and EDA.

I was not satisfied with this accuracy, so I decided to use Deep Learning to solve this problem.